<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> This repository presents an agentic Machine Learning Engineer designed to train state-of-the-art ML models. It leverages the Claude Code SDK and Google ADK, demonstrating the capabilities of Claude Scientific Skills for ML tasks, offering a simplified approach for users to build and train models.

How It Works

The core of the project is an agent that utilizes the Claude Code SDK and Google ADK to orchestrate ML model training. It integrates a collection of "Claude Scientific Skills," which are pre-built tools and workflows for specialized ML tasks. These skills are made accessible within a sandboxed Python environment, managed via an ADK web interface, enabling the agent to leverage advanced scientific libraries and workflows seamlessly.

Quick Start & Requirements

Installation requires Python 3.13+ and the uv package manager. Dependencies are installed via uv sync. An .env file is mandatory, containing OPENROUTER_API_KEY and AGENT_MODEL. The start.py script automates the setup: creating a sandbox, configuring a Python virtual environment with ML packages (PyTorch, transformers, scikit-learn), copying the .env file, and launching the ADK web interface at http://localhost:8000. Manual setup of the sandbox or ADK web interface is also possible via python -m karpathy.utils and adk web, respectively. Datasets or scripts must be manually added to the sandbox directory